X_train, X_test, y_train, y_test, ind_train, ind_test = load_data(full=False)
clf = GradientBoostingClassifier(n_estimators=500, max_depth=6,
learning_rate=0.1, max_features=256,
min_samples_split=15, verbose=3,
random_state=13)
print('_' * 80)
print('training')
print
print clf
clf.fit(X_train, y_train)
if y_test is not None:
from sklearn.metrics import auc_score
print clf
y_scores = clf.decision_function(X_test).ravel()
print "AUC: %.6f" % auc_score(y_test, y_scores)
if generate_report:
from error_analysis import error_report
data = np.load("data/train.npz")
X = data['X_train']
X_test_raw = X[ind_test]
error_report(clf, X_test_raw, y_test, y_scores=y_scores, ind=ind_test)
np.savetxt("gbrt3.txt", clf.decision_function(X_test))
from transform import SpectrogramTransformer
from ranking import RankSVM
from ranking import SVMPerf
from ranking import RGradientBoostingClassifier
import IPython
data = np.load("data/train_small.npz")
X = data["X_train"]
y = data["y_train"]
clf = LinearSVC(C=1e-5, tol=0.001, loss='l1', dual=True)
clf = Pipeline(
steps=[('spectrogram',
SpectrogramTransformer(
NFFT=256, noverlap=0.5, dtype=np.float64)), ('svm', clf)])
ind = np.arange(X.shape[0])
X_train, X_test, y_train, y_test, ind_train, ind_test = train_test_split(
X, y, ind, test_size=0.5, random_state=42)
clf.fit(X_train, y_train)
from error_analysis import error_report
error_report(clf, X_test, y=y_test, ind=ind_test, spec_func=None)
from sklearn.pipeline import Pipeline
from transform import SpectrogramTransformer
from ranking import RankSVM
from ranking import SVMPerf
from ranking import RGradientBoostingClassifier
import IPython
data = np.load("data/train_small.npz")
X = data["X_train"]
y = data["y_train"]
clf = LinearSVC(C=1e-5, tol=0.001, loss="l1", dual=True)
clf = Pipeline(
steps=[("spectrogram", SpectrogramTransformer(NFFT=256, noverlap=0.5, dtype=np.float64)), ("svm", clf)]
)
ind = np.arange(X.shape[0])
X_train, X_test, y_train, y_test, ind_train, ind_test = train_test_split(X, y, ind, test_size=0.5, random_state=42)
clf.fit(X_train, y_train)
from error_analysis import error_report
error_report(clf, X_test, y=y_test, ind=ind_test, spec_func=None)
